torchscript graph mode quant: remove benchmark filter (#44165)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/44165
Allows convolutions to be quantized if `torch.cudnn.backends.benchmark`
flag was set.
Not for land yet, just testing.
Test Plan:
in the gist below, the resulting graph now has quantized convolutions
https://gist.github.com/vkuzo/622213cb12faa0996b6700b08d6ab2f0
Imported from OSS
Reviewed By: supriyar
Differential Revision: D23518775
fbshipit-source-id: 294f678c6afbd3feeb89b7a6655bc66ac9f8bfbc
